3. WORKLOADS ARE EVOLVING
TRADITIONAL
WORKLOADS
● Typically resides on a single large
physical or virtual Machine
● Cannot tolerate any downtime
● Needs expensive high availability tools
found in VMware vSphere
● Application scales up rather than out
OPENSTACK TORONTO 3 | STEPHEN GORDON
CLOUD
FUNCTIONS
● Workload resides on multiple Virtual
Machines
● Tolerates VM failure – if one fails, another
quickly replaces it
● Fault tolerance often built into workload
● Application scales out rather than up
4. WHY OPENSTACK
● Brings public cloud-like capabilities into your datacenter
● Provides massive on-demand (scale-out) capacity
● 1,000's → 10,000's → 100k's of VMs
● It's OPEN!
● Provides flexibility to customize and interoperate
● Open APIs for interacting with interchangeable
backends
● Community development = higher “feature velocity”
● Features and functions you need, faster to market over
proprietary software
OPENSTACK TORONTO 4 | STEPHEN GORDON
5. OPENSTACK MISSION STATEMENT
“To produce the ubiquitous Open Source Cloud
Computing platform that will meet the needs of
public and private clouds regardless of size, by
being simple to implement and massively
scalable.”
OPENSTACK TORONTO 5 | STEPHEN GORDON
6. OPENSTACK CIRCA “ICEHOUSE”
CLOUD INFRASTRUCTURE FOR CLOUD WORKLOADS
● Modular architecture, designed to easily scale out
● Based on (growing) set of core services
OPENSTACK TORONTO 6 | STEPHEN GORDON
7. TROVE
● OpenStack Database-as-a-Service (Trove)
● Provides scalable and reliable Cloud Database as a
Service provisioning functionality
● Supports relational and non-relational database engines
● Provision and manage multiple database instances as
needed
● API supports JSON and XML to provision and manage
instances
OPENSTACK TORONTO 7 | STEPHEN GORDON
9. OPENSTACK SUMMIT
● Six monthly User and Developer conference.
● Nov 2013 – “Icehouse” summit in Hong Kong.
● May 2014 – “Juno” summit in Atlanta.
● Nov 2014 – “Kilo” summit in Paris.
● General track provides venue for traditional presentations
on user stories, new features, and vendor solutions.
● Developer track provides less structured slots for
discussing features and roadmap for coming release.
OPENSTACK TORONTO 9 | STEPHEN GORDON
10. JUNO SUMMIT RE-CAP
● Held at Georgia World Congress Center in Atlanta in
May
● ~4,500 attendees (~3,500 in Hong Kong 6 months prior)
10 OPENSTACK TORONTO | STEPHEN GORDON
11. TECH PREVIEW: TROVE
● OpenStack Database-as-a-Service (Trove)
● Provides scalable and reliable Cloud Database as a
Service provisioning functionality
● Supports relational and non-relational database engines
● Provision and manage multiple database instances as
needed
● API supports JSON and XML to provision and manage
instances
*Tech Preview features are subject to change in GA
OPENSTACK TORONTO 11 | STEPHEN GORDON
12. JUNO SUMMIT RE-CAP – SUPERUSERS
● Increased visibility of “superusers”
● Keynotes including content from:
● AT&T
● Sony
● Wells Fargo
● ...and others
● Operators track in the design summit
● Neutron vs nova-network
● Upgrades
● Launch of http://superuser.openstack.org/
12 OPENSTACK TORONTO | STEPHEN GORDON
15. JUNO SUMMIT RE-CAP – NFV
● Aim to decouple network functions from physical
infrastructure while maintaining performance.
● Increased presence from Communication Service
Providers (CSPs), Network Equipment Providers
(NEPs) etc.
● Formation of NFV subgroup to gather and work on
requirements.
● Similar to “win the enterprise” working group
launched at the Juno summit as well.
15 OPENSTACK TORONTO | STEPHEN GORDON
16. JUNO RELEASE SCHEDULE
● Feature proposal freeze
● Sept 21 – Today!
● Juno-3 release and Feature Freeze
● Sept 4
● Release candidates
● Sep 25 onwards
● Project release
● Oct 16
16 OPENSTACK TORONTO | STEPHEN GORDON
17. JUNO RELEASE PROJECTS
● Integrated:
● Sahara – Formerly Savanna - Big Data service
● Incubated:
● Ironic – Baremetal Hypervisor Driver
● Zaqar – Formerly Marconi, multi-tenant cloud
messaging service like Amazon SQS
● Designate – DNS-as-a-Service
● Barbican - Secure storage, provisioning and
management of secrets.
● Applied: Manila – Filesystem-as-a-Service
17 OPENSTACK TORONTO | STEPHEN GORDON
18. INTEGRATED: SAHARA
● OpenStack Data Processing (Sahara)
● Provisioning and management of Hadoop clusters
● Help identify and improve utilization of unused compute
power from general purpose OpenStack IaaS cloud
● Pluggable system of Hadoop installation engines for
different distros
● Predefined templates of Hadoop configurations with
ability to modify parameters.
OPENSTACK TORONTO 18 | STEPHEN GORDON
19. TECHNICAL COMMITEE FOCUS AREAS
● Neutron feature parity with nova-network
● Migration strategy for moving between the two
(live/cold)
● Scalability – multi-host versus distributed virtual
router
● Test coverage in Tempest
● Retrospectively was integrated too early, policy
changes since applied.
● Which leads to...
19 OPENSTACK TORONTO | STEPHEN GORDON
20. TECHNICAL COMMITEE FOCUS AREAS
● Heat and Ceilometer gap coverage
● Updated integration requirements being applied
retrospectively to integrated projects.
● Scaling issues with both projects in some scenarios.
● Not abstraction layers in the same fashion as some
of the other projects (e.g. Nova and Neutron).
● Documentation coverage improving.
20 OPENSTACK TORONTO | STEPHEN GORDON
21. BOARD FOCUS AREAS
● “Win the Enterprise” working group
● “Engaging hidden influencers” effort
● Both aim to determine how non-developers
effectively contribute to and collaborate on
OpenStack.
● DefCore – Attempt to define what is core and in turn
how the OpenStack trademark can be used by
vendors.
21 OPENSTACK TORONTO | STEPHEN GORDON
22. ARCHITECTURE DESIGN GUIDE
● 12 writers over 5 days @ Vmware HQ in Palo Alto
● Compute, Storage, Network focused architectures
among others.
● Apache License 2.0
● On-line:
● http://docs.openstack.org/arch-design/content/
● Print:
● http://www.lulu.com/ca/en/shop/openstack-foundation/openstack-22 OPENSTACK TORONTO | STEPHEN GORDON
23. OPENSTACK JUNO (TENTATIVE)
● Ironic driver for Nova, replaces nova-baremetal.
● SR-IOV support
● Extend PCI passthrough support for SR-IOV
OPENSTACK TORONTO 23 | STEPHEN GORDON
24. OPENSTACK JUNO
● Scheduler NUMA awareness
● Extend compute driver to track NUMA nodes
● Aim to:
● Ensure colocation of guest CPU and RAM (CPU only initially).
● Avoid floating guest CPU and RAM across nodes .
● Enable intelligent scheduling in guest by exposing topology.
OPENSTACK TORONTO 24 | STEPHEN GORDON
25. OPENSTACK JUNO (TENTATIVE)
● ML2 as the standard for Neutron plugins.
● ML2 was introduced in Icehouse.
● Traditional plug-ins deprecated for removal in Juno.
● Provides more freedom for heterogeneous
environments.
● Distributed virtual router (DVR).
● Further improvements to IPv6 support.
OPENSTACK TORONTO 25 | STEPHEN GORDON
Editor's Notes
IT operations are being challenged more than ever from various ends of an organization, each with very different types of application and workload requirements.
The result…IT operations are straining to meet these new application and workload demands, using their existing traditional infrastructure.
In many cases users are seeking the elasticity and dynamic growth they need, outside of their IT team, creating an uncontrolled “shadow IT” problem.
Very quickly, the benefits of a private cloud are becoming apparent to IT operations, as they seek the right cloud solutions to meet their business demands.
...And these new user demands are driving new application development to be more elastic and dynamic and uniquely designed for use in the cloud. Many advantages to this cloud-enabled workload can be rapidly altered and improved, and the architecture they reside on is significantly cheaper if deployed correctly.Traditional Workloads:- often reside on a single, large, stateful VM with several vCPU, lots of vRAM, storage inside VM, etc. - have a high SLA, and cannot tolerate any downtime—if that single VM goes down the enterprise has the potential to lose real dollars.
- it needs proper care and maintenance, requires enterprise virtualization features to keep VMs highly available, and must be very carefully monitored with frequent troubleshooting
- Scales upward-- as app/user demand grows, the VM gets bigger
Cloud workloads are very different
Workload runs on many small, stateless VMs, each typically small VMs or even container where: vCPU, vRAM, storage separate)
Applications tolerate failure of VMs – If one VM fails, another quickly replaces it
Fault tolerance often built into workload – no longer require expensive resiliency tools provided by Virtualization
Scales outward, rather than up – as app/user demand grows, add more VMs
But what IS OpenStack exactly?
OpenStack is actually a series of several independently developed services that comprise the sub-projects that work together to form a cloud framework.
The framework is intentionally designed to be modular, which provides for massive scale-out capabilities for the entire framework. As user/application demands grow, administrators can simply add new service nodes as needed. The system is designed to scale to support thousands and tens-of-thousands VMs.
Each service, comprise a set of “core” services that make up the framework, and are updated every 6 months with new projects and services.